Ok, maybe I should get more specific. To start with, here's one of the claims of subjectivism Self claims is false:
Degradation effects exist in amplifiers that are unknown to engineering science, and are not revealed by the usual measurements.
In fact, this statement is obviously true. It's essentially an axiom of the scientific method. It's hubris to assume that we've uncovered EVERY distortion that nature put into the amplifier, so of course, there will be some we haven't discovered yet. The burden of proof is therefore on him to show that these unmeasurable distortions don't change the subjective perception of quality. Instead, he assumes that they simply don't exist.
So, if Doug Self has assumed from the beginning that there is no such thing as a distortion mechanism that doesn't show up on his Audio Precision System 2 or whatever, why should we treat it as significant that he doesn't find any?
For instance, here's what he says in his rebuttal to the claim that amplifiers can't be completely characterized with just sine waves:
You must remember that an amplifier has no perspective on the signal arriving at its input, but literally takes it as it comes.
That's quite false. Since amplifiers are not memoryless, their internal state DOES depend on the past history of the music, and since they're nonlinear, other parameters like the gain and distortion spectrum do depend on the previous signal. Sine waves are an easy test of an amplifier because they're
periodic, and thus admit a steady-state solution in which all transient behavior has died out.
Slew rate limiting is an obvious example: if you crank in a step that causes slew rate limiting to kick in, other signals will be silenced entirely until the ramp is done, since the input stage is saturated.
This interaction between linear and nonlinear effects is also the crux of what peufeu is talking about, and his research (along with Lavardin's) clearly shows a potential mechanism for audible distortion that doesn't show up on a THD spectrum. Listening tests (maybe not blind ones) show a very obvious difference from this new effect.
Now, I've never heard a "blameless" amplifier so it may be that its distortion is too low to begin with for any of this to matter, but it does not change the fact that Self's reasoning is wrong.